15

Assembly Sequence Planning Using Neural Network Approach

Cem Sinanoglu and Huseyin Riza Borklu

1. Introduction The competition between manufacturing firm’s makes it necessary that those firms must supply highly quality goods with shorter time and chipper for survive in the international market. The intensive research in this field aims at augmenting methods and tools for product development and manufacturing. By the use of new and efficient methods it is possible to shorten the time from design to manufacturing and reduce the mistakes originating from humans. Therefore, full automation in the manufacturing processes can be accomplished. An assembly can be defined as the overall function of individual parts after joining each other’s, each of which has an independent function. It is possible to divide an assembly into various subassemblies depending on its complexity levels (Pahl & Beitz, 1988). Although intensive research efforts in the field of assembly sequence planning, there are still some problems to be solved. For example, there exist some shortcomings to support for full automatic assembly sequence planning or to obtain assembly plans for large systems (Singh, 1997). By the development of efficient assembly planning systems it is possible to provide some advantages in both areas: • • CAD automation as well as Manufacturing performed by the use of robots

Open Access Database www.i-techonline.com

In order to develop a system that addresses computer-aided assembly sequence planning, these issues should be taken into consideration: • The connectivity structure of parts and/or subassemblies that are used in the assembly system

Source: Manufacturing the Future, Concepts - Technologies - Visions , ISBN 3-86611-198-3, pp. 908, ARS/plV, Germany, July 2006, Edited by: Kordic, V.; Lazinica, A. & Merdan, M.

437

438

Manufacturing the Future: Concepts, Technologies & Visions

• •

If they have connectivity properties, then what are the theatrical number of their different connection ways Finally the choice of the most optimum assembly process among various alternatives

Basic approach for finding the most suitable assembly sequences is to represent assembly system in a space where it is possible to explore all different assembly sequences. Thus, some criterion may be used to obtain these sequences. Then optimum assembly sequence can be selected by the application of some other criterion. This criterion may include: the number of tool changes, part replacement and clamping on the jigs, concurrency in operations, reliable subassemblies, etc (Kandi & Makino, 1996). A new approach may be stated as a rank evaluation from new scanning factor. But this may not exactly show the differences between operation time and costs. The initial assembly planning system was inquired to user data necessity for operation and it was formed to assembly sequences with given data. It was worked with the user interaction (De Fazio, 1987, Homem de Mello & Sanderson, 1991). The later work concentrated on assembly sequence planning systems based on geometric reasoning capability and full automatic (Homem de Mello, 1991). Then, the system included some modules that allow assembly sequence plans to be controlled and to be tested. Since many assembly sequences share common subsequences, attempts have been made to create more compact representations that can encompass all assembly sequences. Therefore, the works in this field are graph-based approaches to represent all feasible assembly sequences (FAS ) . This paper presents a neural-network-based computational scheme to generate feasible assembly sequences for an assembly product. The inputs to the networks are the collection of assembly sequence data. This data is used to train the network using the Back propagation (BP ) algorithm. The neural network model outperforms the feasible assembly sequence-planning model in predicting the optimum assembly sequences. 2. System Definition This assembly planning system is used to assembly’s connection graph ( ACG ) for the representation product. Parts and relations among these parts are represented by this graph. Contact relations between parts are supplied by scan-

Assembly Sequence Planning Using Neural Network Approach

439

ning module. Scanning module scans various assembly views of any product whose assembly view is produced by CAD packet program in the computer environment and determines to contact and interference relations between parts (Sinanoğlu & Börklü, 2004).

Figure 1. Block diagram of proposed assembly planning system

Moreover. whether corresponding to assembly states or not. binary vectors represent assembly states. Artificial Neural Networks An artificial neural network (or simply a neural network-NN) is a biologically inspired computational structure. Due to parallel learning structure of the network. vector representations corresponding to assembly states are determined. 1990). Assembly states supplying constraints are settled down in the nodes of directed graph hierarchically by the system. subassembly and geometric feasibility constraints. Any path from the root node to terminal node in the directed graph corresponds to feasible assembly sequence. The connection weights change during learning (training). . 2005). Therefore. Training and a recall algorithm are associated to every NN.440 Manufacturing the Future: Concepts. The feasible assembly states and assembly sequences are represented by a directed graph (Homem de Mello & Arthur. Some of the assembly states cannot take part in a feasible assembly sequence. are produced by the system. which consists of processing elements (neurons) and connections between them with coefficients (weights) bound to the connections. In the assembly planning system. the proposed neural network has superior performance to analyze these systems. The system constitutes ACG of product to be assembled. The determination of the assembly states not corresponding to feasible assembly sequence is achieved with the analysis module. By evaluating ACG and binary vector representations simultaneously with the scanning module. NN are also called connectionist models because of the main role of the connections the weights are the result of the training phase. The optimum one is selected from the feasible assembly sequences with optimization module. The neural network approach has been employed for analyzing feasible assembly sequences for sample product. interference relations are also accumulated in a form of interference matrices for determination of geometric feasibility of assembly states later (Sinanoğlu & Börklü. by applying Boolean operators on the elements of the contact matrices according to certain rules. all binary vector representations. The analysis module controls all assembly states according to stability. they are the "memory" of the structure. 1. Block diagram of assembly system is shown in Fig. Boolean operators apply to the elements of interference matrices determination of geometric feasibility. 3. Technologies & Visions These relations are formed as a matrix.

. which calculates the aggregated net input Signal U to the neuron (this is usually a summation function) An activation (signal) function. wn An input function ( f ) ... bound to the inputs w1 .. x 2 . x n Weights.. 2(b)): • • • • • • Inputs x1 . Schematic view of a real neuron . The connections between neurons are realized in the synapses.... An artificial neuron is defined by (Fig.Assembly Sequence Planning Using Neural Network Approach 441 A structure of a typical biological neuron is shown in Fig. which calculates the activation Level of the neuron: O = g (U ) Figure 2(a)... 2(a). It has many inputs (in) and one output (out). w2 ..

The connections can represent the current weight values for each weight. 2(c) shows the currently loaded network. output and hidden layers. The neural network consists of three layer. Technologies & Visions Figure 2(b) Schematic representation of the artificial neural network Fig. the rightmost being the output layer. circles depict the neurons.442 Manufacturing the Future: Concepts. which are input. The input and outputs data are used as learning and testing data. Figure 2(c) Currently loaded network . Triangles represent the bias for each neuron. Squares represent input nodes.

η is the learning rate.Assembly Sequence Planning Using Neural Network Approach 443 The most important and time-consuming part in neural network modeling is the training process. In this study. Back propagation is a minimization process that starts from the output and backwardly spreads the errors (Canbulut & Sinanoğlu. 2004). Logistic function is as follows. and α is the momentum term. Back propagation algorithm is used training algorithm for proposed neural networks. logistic function and back-propagation learning algorithm are employed to train the proposed NN. In some cases the choice of training method can have a substantial effect on the speed and accuracy of training. ∂y =1 ∂x (5) Training and structural parameters of the network are given in Table 1. ∂E (t ) + α∆wij (t − 1) ∂wij (t ) ∆wij (t ) = −η (1) where. The best choice is dependent on the problem. y = f (x ) = x (3) (4) Its derivative is. y = f ( x) = 1 1 + e−x (2) Its derivative is. In this study. The weights are updated as follows.(1 − x ) ∂x The linear function is. ∂y = y. and usually trial-and-error is needed to determine the best method. Linear function is taken for input layer. . the logistic function is used to hidden layers and output layers.

Technologies & Visions Proposed Neural Network η 0 .. The connections or edges defining relationships between parts or nodes can be stated as: c1 between parts {a} and {b} . {b} . These parts are represented respectively by the symbols of {a}. In order to modelling assembly system.. Modeling of Assembly System An assembly is a composition of interconnected parts forming a stable unit. The assembly process consists of a succession of tasks. and that after parts have been put together. The process starts with all parts separated and ends with all parts properly joined to form the whole assembly.. For this particular situation. c3 between parts {c} and {d }. c. c 2 . c 4 . we will take a sample assembly shown as exploded view in Fig. and C = {c1 .1 µ 0 nI nH 10 nO N 500000 AF log istic 1 4 Table 1.. For the current analyses. C . c 2 . it is used ACG whose nodes represent assembling parts and edges represent connections among parts. the remain together until the end of the assembly process. {c} and {d }..444 Manufacturing the Future: Concepts. in which P = {p1 .. 3.. Each node in P corresponds to a part in the assembly. d } and the set of the connections as C = {c1 . The sample assembly is a pincer consisting of four components that are: bolt. left-handle. . b. c 2 between parts {a} and {d }.. and there is one edge in C connecting every pair of nodes whose corresponding parts have at least one surface contact. p N } is the set of nodes. right-handle and nut. In order to explain the modeling of assembly system approach better way used for this research. c5 } . Training and structural parameters of the proposed network 4. c L } is the set of edges. an assembly can be represented by a simple undirected graph P. each of which consists of joining subassemblies to form a larger subassembly. the connection graph of assembly has the set of the nodes as P = {a. Due to this assumption. it is assumed that exactly two subassemblies are joined at each assembly task. p 2 . c 4 between parts {a} and {c} and finally c5 between parts {b} and {c}.

To determine these relations. The pincer assembly system 4. b] element of B contact matrix in i direction is defined to whether there exists any contacts or not between parts {a} and {b} for the related direction and the corresponding matrix element may have the values of (1) and (0) . If there is. respectively.1 Definition of Contact Matrices and ACG The contact matrices are used to determine whether there are contacts between parts in the assembly state. Then it is examined that whether this part has a contact relation in any i axis directions with other parts. The elements of these matrices consist of Boolean values of true (1) or false (0 ) . that relation is defined as true (1) . [a. the first part is taken as a reference. the assembly’s parts are located to rows and columns of the contact matrices. For the construction of contact matrices.Assembly Sequence Planning Using Neural Network Approach 445 Nut (d) Right-Handle (c ) Left-Handle (b) Bolt (a) Figure 3. B}. . These matrices are represented by a contact condition between a pair of parts as an {A. For example. else that is defined as false (0 ) . The row and column element values of contact matrices in the definition of six main coordinate axis directions are relations between parts and that constitutes a pincer assembly. Contact matrices are square matrices and their dimensions are 4 × 4 for pincer.

assembly view of pincer system was used. b} sequenced pair of parts. in order to get contact matrices in the direction of Cartesian coordinate axis. an [∨ : Or ] operator is applied to these parts. But it is . Contact matrices of the pincer assembly system are also shown in Fig.446 Manufacturing the Future: Concepts. in the manner of {a. 4. For instance. Contact matrices and their graph representations In this system. contact conditions are examined in both part’s sequenced directions. To construct ACG . it is sufficient to determine contacts related sequenced direction so that its contact in any direction. The connection graph can be obtained from contact matrices. Technologies & Visions a b c d ⎡0 ⎢1 By = ⎢ ⎢1 ⎢ ⎣1 1 1 1⎤ 0 0 0⎥ ⎥ 0 0 0⎥ ⎥ 0 0 0⎦ a b c d ⎡0 ⎢1 Bz = ⎢ ⎢1 ⎢ ⎣1 1 1 1⎤ 0 0 0⎥ ⎥ 0 0 0⎥ ⎥ 0 0 0⎦ a b B− x c d ⎡0 ⎢0 =⎢ ⎢0 ⎢ ⎣0 1 0 0⎤ 0 1 0⎥ ⎥ 0 0 1⎥ ⎥ 0 0 0⎦ a b B− y c d ⎡0 ⎢1 =⎢ ⎢1 ⎢ ⎣1 1 1 1⎤ 0 0 0⎥ ⎥ 0 0 0⎥ ⎥ 0 0 0⎦ a b B− z c d ⎡0 ⎢1 =⎢ ⎢1 ⎢ ⎣1 1 0 0 0 1 0 0 0 1⎤ 0⎥ ⎥ 0⎥ ⎥ 0⎦ a b c d ⎡0 ⎢1 Bx = ⎢ ⎢0 ⎢ ⎣0 0 0 0⎤ 0 0 0⎥ ⎥ 1 0 0⎥ ⎥ 0 1 0⎦ Figure 4. 2004). Due to this reason. These matrices were automatically constructed (Sinanoğlu & Börklü.

5. There is no contact between the left-handle and the nut. c} . 5 shows the pincer graph of connections. b} sequenced pair of parts is supplied to at least one contact condition in the related direction of (0 ∨ 1 ∨ 1 ∨ 1 ∨ 1 ∨ 1 = 1) . 5. then there should be edges between corresponding nodes of the ACG . If these values are (1) for every sequenced pair of parts. a} pair of parts is also supplied to at least one contact in the related direction of (1 ∨ 1 ∨ 1 ∨ 0 ∨ 1 ∨ 1 = 1) . d }. where there is no edge between parts {b} and {d }. {d . {c. Table 2 shows contact relations regarding {a. Therefore. {b. Fig. If similar method is applied to other pairs of parts: {a. It has four nodes and five edges (connections). the results should be (1) . c} and {c. it should be searched contact relations for these pairs of parts. {c. b} and {b. a}. d }. {d . Contact relations of {a. {a. b}. Therefore. For this purpose. b} and {b. c1 ⇒ (a ÷ b) ab b a ∨ : Or ⇒ x 0 y 1 1 1 z 1 1 −x −y 1 1 1 0 −z 1 1 ∧ : And ⇓ 1 1 1 1 1 Table 2. these parts do not have any contact relations. a} pairs of parts. An (∧ : And ) operator is applied to these obtaining values. a}. there is an edge between parts in the ACG . b} . there are edges between these pairs in ACG . • {a.Assembly Sequence Planning Using Neural Network Approach 447 also necessary contacts in any direction for inverse sequenced pairs of parts in the ACG . {b. c}. This connection states an edge in the ACG shown in Fig. The graph representation of this situation is shown in Fig. a} Sequenced pair of parts To investigate whether there is an edge between {a} and {b} in ACG or not. There- . every pair of parts must be determined. {a. Because. these parts have at least one contact in each part sequenced direction. {b. a} pairs of parts In this table.

Therefore. For example. x L }) . .. it is proved that it is supplied to one connection between two part in ACG not all contacts between them are established in every direction. the graph of connections does not include an edge connecting the nodes corresponding to the left-handle and the nut. Moreover. By the use of the contact matrices and applying some logical operators to their elements. the established connections in binary vectors and ACG are utilised together. in the developed approach an L-dimensional binary vector can represent a state of the process (x = {x1 . The graph of connections for four-part pincer assembly 5. Otherwise. Technologies & Visions fore. The configuration of parts is given by the contacts that have been established. every binary vector representations are not corresponding to an assembly state. In order to determine assembly states.448 Manufacturing the Future: Concepts.. it would have a value of false (0 ) .. c1 a b c4 c5 c2 c c3 d Figure 5. x2 .. the i th component xi would have a value of true (1) if the i th connection were established at that state. Determination of Binary Vector Representation and Assembly States ( ASs ) The state of the assembly process is the configuration of the parts at the beginning (or at the end) of an assembly task. Based upon the establishment of the connections. the elements of these vectors may have the values of either (1) or (0 ) at any particular state of assembly task. Elements of these vectors define the connection data between components.

. if the first task of the assembly process is the joining of the bolt to the left-handle. [01000]. The purpose of that it is classified to binary vectors according to the number of established connections. For instance. 6 that it does not necessary to establish any connection so that c1 connection between part {a} and {b} is establish. the second state of the assembly process can be represented by [FTFFF ] .Assembly Sequence Planning Using Neural Network Approach 449 There are five edges in the example ACG . c3 . The third element represents c1 connection between right-handle and bolt and c3 connection between right-handle and nut. first of all binary vector representations must be produced. c5 ] . In order to determine whether the vector is a state or not. Because of that. Therefore. There are thirty-two different binary vectors. Therefore. 3. [TTTFF ]. vector representations corresponds to assembly states must be determined. 3 can be represented by binary vector [FFFFF ] whereas the final state can be represented by [TTTTT ] . Table 3 shows vector representations for pincer assembly in Fig. And then it is required that establishing connections must be determined to established connections by ACG . While some of them correspond to assembly state. c2 . In the developed planning system. To form assembly sequences of pincer system. it must be taken into consideration established connections in vector representation. [11111]) The first element of this list represents the initial state of the assembly process. [11100]. an assembly sequence for pincer system can be represented as follows: ([FFFFF ]. the elements of vectors are five and the 5-dimensional binary vector of can represent that [c1 . [FTFFF ]. some of them are not. If the first task of the assembly process is the joining of the bolt to nut. [10000] vector is an assembly state. vectors only one established connection form assembly state. The second element of the list shows the second connection c 2 between bolt and nut. the initial state of the assembly process for the product shown in Fig. For example. The last element of the list is [11111] and it means that every connection has been established. It is seen in Fig. For instance. c4 . the second state of the assembly process can be represented by [10000] . [TTTTT ]) ([00000].

Technologies & Visions LEVEL 0 LEVEL 1 11000 10100 10000 10010 10001 LEVEL 2 11100 11010 11001 10110 10101 10011 11100 LEVEL 3 11110 11101 11011 10111 11000 01000 01100 01010 01001 11010 11001 01110 01101 01011 11100 11110 11101 11011 01111 10100 00000 00100 01100 00110 00101 10110 10101 01110 01101 00111 11010 11110 11101 10111 01111 11111 10010 00010 01010 00110 00011 10110 10011 01110 01011 00111 11001 11110 11011 10111 01111 10001 00001 01001 00101 00011 10101 10011 01101 01011 00111 11101 11011 10111 01111 Table 3.450 Manufacturing the Future: Concepts. Hierarchical levels of binary vector representations for pincer assembly system .

[00101 ].Assembly Sequence Planning Using Neural Network Approach 451 Moreover.Tp corresponds to directed graph of pincer system. But this connection has not been established in [10001] . A path in the directed x p . 1991). connections of c1 between {a} and {b} . [10011 ]. C = {c1 . {b}. x p . 6). c. c2 . [00000 ]. [01110 ]. These are. c5 between {b} and {c} have been established (1). [10001] vector is not an assembly state. in the [10001] vector. c. Productions and Representation of Assembly Sequences Given an assembly whose graph of connections is P.Tp graph of feasible assembly sequences whose initial node is Θ I = {{a}. It has been necessary to establish c 4 connection between {a} and {c} so that these connections have been established (Fig. C . {d }} and whose terminal node are Θ F = {{a. b. [00100 ]. [10100 ]. [01000 ]. and T p is the assembly’s set of feasible state transitions. c3 . d }} . d } is the assembly’s set of parts or set of nodes. c5 } is the assembly’s set of connections or set of edges. c1 a b c4 c5 c d Figure 6 c1 . b. Vector representations of these sets are [00000] and [11111] respectively. [11000 ]. [10000 ]. {c}. a directed graph can be used to represent the set of all assembly sequences (Homem de Mello & Lee. x p is the assembly’s set of stable states. c5 and c 4 connections in [10001] vector There are thirteen assembly states in pincer assembly system. [00010 ]. In the pincer assembly. [11111 ] 6. . T p in which. The directed graph of feasible assembly sequences of an assembly whose set of parts is P is the directed graph x p . For instance. P = {a. [01001 ]. [00001 ]. some of vectors do not correspond to an assembly state. c4 .

Interference matrices and their graph representations for pincer assembly . In order to form a subassembly of a set of partition. it is not a set of partition contains a pair of part has not contact relation in the ACG . Therefore. stability and geometric feasibility constraints. In this study. Technologies & Visions Assembly states not correspond to feasible assembly sequences must eliminate by some assembly constraints. Because of that it is not supplied to subassembly constraint set partitions contains {b. three assembly constraints are applied to assembly states. a b c d ⎡1 ⎢0 Ax = ⎢ ⎢0 ⎢ ⎣0 ⎡1 ⎢0 Az = ⎢ ⎢0 ⎢ ⎣0 ⎡1 ⎢0 =⎢ ⎢0 ⎢ ⎣0 1 1 1⎤ 1 1 1⎥ ⎥ 0 1 1⎥ ⎥ 0 0 1⎦ 0 0 0⎤ 1 0 1⎥ ⎥ 0 1 1⎥ ⎥ 1 1 1⎦ 0 0 0⎤ 1 0 1⎥ ⎥ 0 1 1⎥ ⎥ 1 1 1⎦ a b c d ⎡1 ⎢0 Ay = ⎢ ⎢0 ⎢ ⎣0 ⎡1 ⎢1 A− x = ⎢ ⎢1 ⎢ ⎣1 ⎡1 ⎢0 A− z = ⎢ ⎢0 ⎢ ⎣0 0 0 0⎤ 1 0 1⎥ ⎥ 0 1 1⎥ ⎥ 1 1 1⎦ 0 0 0⎤ 1 0 0⎥ ⎥ 1 1 0⎥ ⎥ 1 1 1⎦ 0 0 0⎤ 1 0 1⎥ ⎥ 0 1 1⎥ ⎥ 1 1 1⎦ a b a b c d c d a b a b A− y c d c d Figure 7. The second constraints is stability. The subassembly constraint defines feasibility of subassembly of set of partitions to established connections in assembly states.452 Manufacturing the Future: Concepts. in pincer assembly bolt and left-handle has not contact relations. All one-part subassemblies are stable. These are subassembly. d } set of partition. A subassembly is said to be stable if its parts maintain their relative position and do not break contact spontaneously.

it is taken into consideration interference while the reference part is moving with another part along with related axis direction. whether binary vector representations corresponds to assembly states are geometrically feasible or not. These matrices are also formed automatically from various assembly views. At this operation. Therefore. But the movement of left-handle along with {+ x} axis does not interfere any parts (Fig. it must be utilise established connections and that is joining pairs of part. In order to. In the determination of geometric feasibility. interference matrices elements are (0 ) if not are defined as (1) . in the Ax matrice the movement of part bolt is interfered to movement along with {+ x} axis by other parts. For instance. 7). connection of c 2 between part {a} and {d } has been established. • [01000] Assembly State In this assembly state. To determine geometric feasibility of this assembly state. In order to determine whether assembly states are geometrically feasible or not. parts . the first row elements of Ax matrice defined to interference among parts along this axis are (1) . Geometric feasibility of binary vectors correspond to assembly states are determined by interference matrices.Assembly Sequence Planning Using Neural Network Approach 453 The last constraint is geometric feasibility. it is necessary to apply Cartesian product between sequenced pairs of parts which represent established connections and parts which are not in this sequenced pairs of parts. it is applied to elements of interference matrices (∧) and (∨ ) logical operators. different interference tables are obtained and these tables are used to check geometric feasibility. An assembly task is said to be geometrically feasible if there is a collision-free path to bring the two subassemblies into contact from a situation in which they are far apart. In this situation. The elements of interference matrices were taken into consideration interference conditions during the joining parts. This interference relation is illustrated to designate (0 ) value by element of second row and third column in Ax matrice. If it is interference during this transformation motion. Graph representations for the pincer assembly and construction of their interference matrices can be also determined as follows (Fig. 3). it is necessary to applying Cartesian product between sequenced pairs of parts which are representing established connections and parts which are not in this sequenced pairs. In the determination of interference matrices elements.

d } sequenced pair of part represents established connection c 2 . Those are {b} and {c}. [11111] vector is also geometrically feasible. other assembly states of [00100] and [00001] are geometrically feasible. [10100] and [00101] assembly states contain two established connection are geometrically feasible. d )(c) ⇒ (a. [11000]. As a result of (∧) logical operator is (0 ) (1 ∧ 0 = 0 ) . [10011] assembly state contains three connections that are geometrically feasible but [01110] is geometrically unfeasible.454 Manufacturing the Future: Concepts. Similarly. c} and {d . c} pairs Although it is geometrically feasible (1) to disassemble from {a. but [01001] is not geometrically feasible. As a result of Cartesian product is (a. d } and {b} is given as follows. d } to {b} . This result explained that [01000] assembly state is geometrically unfeasible. (a. Moreover. Table 5 shows interference of them. c2 ⇒ (a ÷ d ) ab x y 0 1 z 0 −x 0 −y 0 −z 0 ∨⇒ 1 Table 4. c) . which is between {a. Moreover. c)(d . Technologies & Visions without in established conditions are taken. c2 ⇒ (a ÷ d ) a c d c x y 0 (a c ∧ d c) ⇓ 1 0 0 z 0 −x 0 −y 0 −z 0 1 0 ∨⇒ 0 1 0 1 0 1 0 1 0 Table 5. it is not geometrically feasible (0 ) to disassemble from {a. d } to {c}. b) Table 4 shows interference’s of {a. Cartesian product. b} sequenced pair of part. {a. Interference of {a. d )(b) ⇒ (a. b)(d . but [00010] is geometrically unfeasible. b} sequenced pair of part Another part without parts constituted assembly state is part {c}. Interference relations of {a. The number of nodes is reduced from 15 to 8 in the di- .

The feasible assembly sequences for pincer assemblies are as follows: FAS − I FAS − II FAS − III FAS − IV FAS − V FAS − VI [00000][10000][10011][11111] [00000][10000][10100][11111] [00000][00001][10011][11111] [00000][00001][00101][11111] [00000][00100][00101][11111] [00000][00100][10100][11111] .Assembly Sequence Planning Using Neural Network Approach 455 rected graph by applying assembly constraints. Constrained directed graph for pincer system Fig. The assembly states supplied to these constraints are as follows: [00000][10000][00100][00001][10100][00101][10011][11111] Root [00000] [10000] [00001] [00100] [10011] [10100] [10011] [00101] [00101] [10100] [11111] Terminal node Figure 8. 8 shows the directed graph of feasible assembly sequences after applied constraints. A path in the directed graph of feasible assembly sequences whose initial node is [00000] and terminal node is [11111] corresponds to a feasible assembly sequences for pincer.

The weights and volumes of parts were calculated automatically with a CAD program.3843kg ) . 7.1. Therefore. Optimization of Weight Criterion In order to determine the optimum assembly sequence. After that these costs are used as a reference to different assembly states. The total weight of assembly states in the optimum sequence according to weight criterion can be defined as. all assembly states in an assembly sequence must be taken into consideration. like bolts and nuts must be assembled last (Bunday. It has been obtained from feasible assembly sequences in previous sections. In order to optimize the assembly sequence. left-handle (0. In this section. The weights of parts of the pincer system are as follows: Bolt (0. The difference of weight is multiplied by unit weight value (100 ) . Finally. weight and the subassembly’s degree of freedom. . The heaviest and bulkiest part is selected as a base part and then the assembly sequence continues from heavy to light parts. Calculated weight costs of assembly states in the assembly sequence are compared with reference weights. the total established connection weights of each assembly state in optimum assembly sequence could be determined. and then the total cost of each path from root node to terminal is calculated the minimum cost sequence is selected as an optimum one. the pincer assembly system is taken as an example. Using the weight criterion. two criteria are developed. connective parts. The parts with the least volume. right-handle (0.0092kg ) . at first. in the third assembly sequence for pincer system. determination of the costs of assembly states is necessary to obtain an optimum feasible assembly sequence. 7. For this purpose. i. the left handle is joined to right handle with connection of c5 . Technologies & Visions For example. Optimization of Assembly Sequences Developed assembly planning system can be determined to find the optimum assembly sequence. After that this subassembly is joined by using the bolt with the connections of c1 and c 4 .0163kg ) .3843kg ) and nut (0.456 Manufacturing the Future: Concepts. First certain costs are assigned to edges of directed graph depend on these criteria. an optimization approach is explained. 1984). the nut fixes all parts.e.

7686kg . (n ) is the number of the established connections and (m ) is the order of assembly states. The result is the weight costs of any assembly states (Wc ) . Cwm = ∑ (W ) n i =1 (7) where W is the weight of assembly states. This weight difference (Dw) is multiplied by unit weight value (Uwv = 100) . in the second assembly state of the first feasible assembly sequence ([10000]) . The number of the established connection is n = 1 . the weights of all assembly states in assembly sequences are calculated. This weight is expressed as. c1 connection between {a} and {b} is established. (Owm ) is used as a reference for different assembly states. . Dwm = Owm − Cwm Wc = ∑ ((Wc )m ) = l (8) ∑ ((Ow l i =1 i =1 m l ⎛⎛ ⎞ ⎞ ⎞ ⎞ ⎛ n ⎛ n ⎟ xUwv ⎟ ( ) ( ) W W − − Cwm )xUwv ) = ∑ ⎜ ⎜ ⎟ ⎜ ⎟ ⎜ ∑ ∑ ⎟ ⎜⎜ ⎟ i =1 ⎝ ⎝ ⎝ i =1 ⎠H ⎠ ⎠O ⎝ i =1 ⎠ (9) state is Cw2 = ∑ (W )H = W (c1 ) = 0.Assembly Sequence Planning Using Neural Network Approach Owm = ∑ (W ) n i =1 457 (6) where W is the weight of assembly states. (n ) is the number of the established connections and (m ) is the sequenced of assembly states. 1 i =1 For example. Calculated weights of assembly states in assembly sequences are compared with reference weights. After that. the necessary weight is Ow2 = 0. The required weight of this In the second assembly state.4006kg . In order to determine the optimum assembly sequence. Owm is the required weight of assembly states in the optimum assembly sequence.

It is "25" and this criterion is more important than the other. in the directed graph costs of degree of freedom according to this criterion are calculated as the degree of freedom for each path from initial node to terminal node.368 x100 ≅ 37 The total weight cost of any feasible assembly sequence is expressed as. Wc2 = Dw2 . and so that total cost of assembly sequences can be reduced.368kg . If Dw is multiplied by the unit weight value (Uwv = 100 ) . Technologies & Visions The difference of weight Dw is Dw2 = Ow2 − Cw2 = 0. So degree of freedom between the subassembly parts is low. the minimum cost of the assembly sequence can be selected as an optimum with respect to the degree of freedom criterion. Optimization of Subassembly Degree of Freedom Criterion The subassembly degree of freedom criterion is based on the selection of parts with low degrees of freedom. Udofv ) also used for this criterion. 7. It is a unit cost (unit degree of freedom value. The total degree of freedom (Tdof ) for pairs of parts is the product’s unit cost. it is selected as the lower unit cost according to weight criterion. It determines degree of interference for pairs of parts connections established along the six main directions of the Cartesian coordinate system. the weight cost of ([10000]) will be calculated as follows. DOFc = Tdof × Udofv (11) Therefore.458 Manufacturing the Future: Concepts. DOFct = ∑ DOFc z i =1 (12) . As a result. the assembly of these parts can be done more easily.7686 − 0.4006 = 0. Therefore. The total weight cost of any feasible assembly sequence is expressed as. Wct = ∑Wc z i =1 (10) where z is the total assembly state number of any feasible assembly sequence.2.Uwv = 0.

If this value is multiplied by the Udofv = 25 unit freedom cost. Wc = [(Ow1 + Ow2 + .. in the second assembly state of the first feasible assembly sequence ([10000]) . c1 connection between {a} and {b} is established. the result will be "50" .. + Cwk )]xUwv ⎡⎛ l ⎞⎤ ⎞ ⎛ l = ⎢⎜ ∑ Owk ⎟ − ⎜ ∑ Cwk ⎟⎥ xUwv ⎠⎦ ⎠ ⎝ i =1 ⎣⎝ i =1 (13) DOFc = TdofxUdofv fc = Wct + DOFct (14) (15) is the number of assembly states in the feasible assembly sequence for any product.. c1 ⇒ (a ÷ b) ab b a (l ) x y 0 0 1 0 z 0 0 −x 0 −y 0 0 1 −z 0 0 Table 6.. Therefore. 9 shows feasible assembly sequences and costs of them for the pincer assembly system.. the degree of freedom cost DOFc for [100000000 ] assembly state is "50" . The first and third assembly sequences for the pincer system according to the subassembly degree of freedom criterion have been selected . The total cost of feasible assembly sequence for any product is expressed as a cost function fc . Fig. Degree of freedom between parts {a} and {b} The total degree of freedom for ([10000]) is Tdof = 2 ...Assembly Sequence Planning Using Neural Network Approach 459 where z is the total assembly state number of any feasible assembly sequence. The degree of freedom of this pair of parts is shown in Table 6. The number of the established connection is n = 1 . For example. + Owk ) − (Cw1 + Cw2 + .

() Root [00000] (36). “300” [10100] [11111] Terminal Node Figure 9. In the optimiza- . the sixth assembly sequence is the least preferable sequence in the feasible assembly sequences. The weight costs are in parentheses and the degree of freedom costs are in quotation marks “ ”. “300” [00101] (77). “50” [00001] [00100] (0). “250” [10011] (40).460 Manufacturing the Future: Concepts. 9 shows that the third assembly sequence is optimum "0" weight cost and "300" degree of freedom cost. “250” [10011] (77). “300” [10100] (0). “50” (37). Technologies & Visions with an optimum total cost of "300" . “300” [00101] (40). The weight and degree of freedom costs for pincer system Fig. Moreover. “250” [10000] (0).

[00001]. Therefore. the optimum assembly sequence for pincer system is [00000]. [10011]. Fig. Figure 10(a). [11111] .Assembly Sequence Planning Using Neural Network Approach 461 tion approach. The desired feasible assembly sequences for pincer assembly system Fig. 10(b) (Case 2) is also shows these feasible assembly sequences for neural network approach. . both optimization criteria indicated that assembly sequence is optimum. 10(a) (Case 1) shows the desired feasible assembly sequences for pincer assembly system.

462 Manufacturing the Future: Concepts.6 2 RMS Error 3. RMS E rro r vs Training T im e 5.5 8 2.1 7 4. The feasible assembly sequences for proposed neural networks approach. Technologies & Visions Figure 10 (b).5 2 0.5 5 1.0 0 5 12 00 10 24 00 15 36 00 2 04 80 0 2 56 00 03 0 72 00 35 84 00 40 9 60 0 4 60 80 0 5 12 00 0 E p oc hs Tra ining E rror Te sting E rror Figure 11.6 5 4. The error convergence graph of the case 2 .1 0 2.0 7 1.1 3 3.0 3 0.

c . four-part hinge system and seven-part coupling system have been investigated. Small epoch can be employed for case 2 (51200 epoch). 12 shows this assembly’s exploded views and ACG . As can be seen from the figure. Nut . Handle. Bolt. Plate. d .Assembly Sequence Planning Using Neural Network Approach 463 The error convergence graph of the case 2 is depicted in Fig. b . some sample assembly systems are examined. Some Assembly Case Studies In this work. 8. Among these examples. 11 during the training of the network. the error is suddenly reducing to small values. Nodes of ACG : a . Fig.

Technologies & Visions Nodes of ACG : a . Coupling-I. Washer. b . g .464 Manufacturing the Future: Concepts. Shaft-II. Bolt Figure 12. Nut. f . e . The hinge and coupling systems and their ACG . c . Coupling-II. d . Shaft-I.

In the coupling assembly system. c5 . Figure 13. [000000000 ] and [111111111] are the same as all assembly sequences. the plate and handle are connected with the connection of c1 . [010010000]. [111111111] . And the assembly process is completed with the addition of nut. [010110000]. 13 shows feasible assembly sequences for hinge system. The first one is [00000] and the last is [11111] . The feasible assembly sequences for hinge system For instance. [010000000]. Then using bolt this subassembly is fixed with the connections of c 4 .Assembly Sequence Planning Using Neural Network Approach 465 The assembly sequences of the hinge system contain four different assembly states. in the second assembly sequence. Some of the feasible assembly sequences for coupling system are shown in Table 7. One of the feasible assembly sequences for coupling system is: [000000000]. Fig.

is explained and some assembly systems are examined. and then the washer and nut is joined to last subassembly. Technologies & Visions FEASIBLE ASSEMBLY SEQUENCES FOR COUPLING SYSTEM 000000000 000000000 000000000 000000000 000000000 000000000 000000000 000000000 010000000 010000000 010000000 010000000 010000000 010000000 010000000 010000000 010010000 010010000 010010000 010010000 010010000 010010000 010010000 010010000 011010000 011010010 111010010 011010010 010110000 010111000 010110010 010110001 110011000 111111111 111111111 111111111 111111111 111111111 111111111 111111111 111111111 011010000 011010000 010110000 010110000 010110000 010011000 Table 7. Several assembly sequences of coupling systems The coupling-I is joined to the shaft-I in the first assembly state. the neural predictor would be used as a predictor for possible assembly system applications. 9.466 Manufacturing the Future: Concepts. three assembly sequences for four-part hinge system. the coupling-II and the shaft-II are added in second assembly state. The properties and originalities of the developed system. Conclusions In this chapter. three hundred and seventy three assembly sequences for seven-part coupling system have been investigated. Among these examples. Easy Representation • • • The representation of assembly models by ACG The representation of contact and interference relations between parts by interference matrices The representation of assembly steps by binary vectors . an assembly sequence planning system. As can be depicted from the results. the bolt is used to fix this subassembly. based on binary vector representations. NN application was presented for analyzing assembly sequences on assembly system. six assembly sequences for four-part pincer system. In the third assembly state.

AC. 6. Homem de Mello. 188-199. Vol. Vol. Kluwer Academic Publishers. References Canbulut. ISSN: 1552-3098. 2. 47. CD.Assembly Sequence Planning Using Neural Network Approach 467 Automatic Production • • • • • • • Contact and interference matrices ACG All binary vectors Assembly states Geometric feasibility of assembly states The automatic determination of assembly states that have the conditions of stability and subassembly The automatic determination of feasible assembly sequence plans and optimum assembly one using NN Practical use • • It is possible to obtain appropriate assembly sequence plans after giving various assembly drawings of a product It is understood that the determination of assembly sequence plans that save time and cost after the application of the developed systems to various products 10. No. Homem de Mello. . Homem de Mello. No. ISSN: 15523098. Vol. TL. Computer Aided Mechanical Assembly Planning. 864-872. ISSN: 1552-3098. An Investigation on the Performance of Hydrostatic Pumps Using Artificial Neural Network. & Sinanoğlu. Massachusetts. Vol. LS. 2. & Lee. & Sanderson. (2004). Simplified Generation of All Mechanical Assembly Sequences. F. (1990). C. IEEE Transaction on Robotics and Automation. 6. LS. No. AND/OR Graph Representation of Assembly Plans. De Fazio. & Arthur. LS. JSME International Journal Series C. 640-658. (1991). 3. DE. IEEE Transaction On Robotics and Automation. IEEE Transaction On Robotics and Automation. 7. (1987). (1991). S. ISSN: 1344-7653. A Correct and Complete Algorithm for The Generation of Mechanical Assembly Sequences. 228-240. & Whitney. ISBN: 0792392051.

Vol. Annals of the CIRP. (2004). & Beitz. Pahl. Engineering Design. ISSN: 1726-0604. ISSN: 0956-5515. Singh. (1997). John Wiley & Sons Inc.2. (2005). Assembly Automation. Industrial Lubrication and Tribology. 1. 74-87.. & Börklü. No.R. 56. Sinanoğlu. Vol. & Makino. ISBN: 0-471-58517-3. N. S. ISSN: 0144-5154. 38-52. An Assembly Sequence Planning System for Mechanical Parts Using Neural Network. & Yõldõrõm. International Edition. (1988). Ş. C. Technologies & Visions Kandi. H. Vol. No. The Design Council. 35-39. Analysis of Pressure Variations on Journal Bearing System Using Artificial Neural Network. SpringerVerlag. Systems Approach to Computer Integrated Design and Manufacturing. . ASPEN: Computer Aided Assembly Sequence Planning and Evaluation System Based on Predetermined Time Standard. An Approach To Determine Geometric Feasibility To Assembly States by Intersection Matrices in Assembly Sequence Planning. 543-59. Vol. 1. HR. (1996). AO. Sinanoğlu. G. H.468 Manufacturing the Future: Concepts. Kurban. Sinanoğlu. W. ISBN: 3-540-50442-7. C. C. 25. ISSN: 0036-8792. Journal of Intelligent Manufacturing. No. (2004). 15.. 45. London. & Börklü.

200040.Manufacturing the Future Edited by Vedran Kordic. rapid prototyping. Hotel Equatorial Shanghai No.com/books/manufacturing_the_future/assembly_sequence_planning_using_neural_net InTech Europe University Campus STeP Ri Slavka Krautzeka 83/A 51000 Rijeka.65. Shanghai. InTech. feel free to copy and paste the following: Cem Sinanoglu and Huseyin Riza Borklu (2006). Germany / ARS.). chain management. 2006 Published in print edition July. This interdisciplinary and comprehensive volume. covers a survey of trends in distributed manufacturing. Aleksandar Lazinica and Munir Merdan ISBN 3-86611-198-3 Hard cover. Manufacturing the Future. from technological and organisational point of view and aspects of supply How to reference In order to correctly reference this scholarly work. Available from: work_approach http://www. 908 pages Publisher Pro Literatur Verlag. Yan An Road (West). Croatia Phone: +385 (51) 770 447 Fax: +385 (51) 686 166 www. July. product design process. Vedran Kordic. modern manufacturing equipment. consisting of 30 chapters. quality assurance.intechopen.com InTech China Unit 405. China Phone: +86-21-62489820 Fax: +86-21-62489821 . ISBN: 386611-198-3. 2006 The primary goal of this book is to cover the state-of-the-art development and future directions in modern manufacturing systems. Aleksandar Lazinica and Munir Merdan (Ed. Assembly Sequence Planning Using Neural Network Approach. Office Block. Austria Published online 01.intechopen.